Bonsai 27B
Open-source 27B multimodal model with 1-bit and ternary variants that run locally on a phone or laptop for on-device agentic workflows.
Updated 2026-07-15
Yes. Bonsai 27B is free to use.
Listed pricing: Free — open-source weights under Apache 2.0.
Overview
Bonsai 27B is an open-source, 27-billion-parameter multimodal model family from PrismML, released July 14, 2026, that ships in extreme low-bit variants — including ternary and 1-bit weights — designed to run natively on phones and laptops rather than in a datacenter. The pitch is that you get a model in the 27B class, historically a server-side weight class, small enough to load and run on an iPhone while keeping performance close to a full-precision version.
It's aimed at developers and researchers building on-device and agentic workflows: local assistants, offline multimodal apps, and privacy-sensitive pipelines where sending data to a hosted API is a non-starter. Because the weights are Apache 2.0, teams can fine-tune, quantize further, and ship Bonsai inside a commercial product without licensing friction — the same freedom that made Llama-class and Mistral open weights popular, but at a footprint that targets edge hardware instead of a GPU box.
What sets Bonsai apart is the bet on ternary and 1-bit quantization as a first-class release rather than an afterthought. Most open models ship at full precision and leave community members to produce aggressive quants that lose quality; PrismML is claiming the low-bit variants are the point, positioning Bonsai as the first major 27B-class model engineered to actually fit and run on a phone. Whether the near-full-precision claim holds up under independent benchmarking is the open question — verify the specific quality-vs-size tradeoff against your own eval before betting a product on it.
Is Bonsai 27B free?
Yes. Bonsai 27B is free to use.
What the free tier covers: Entirely free and open-source — weights released under Apache 2.0 with no paid tier from PrismML.
Listed pricing: Free — open-source weights under Apache 2.0.
Pricing on this page has not been re-verified. The entry was last edited on , and no separate pricing check has been run since. Treat the figures as a record of what was published then and confirm on the official site.
Bonsai 27B pricing
| Plan | Price | What's included |
|---|---|---|
| Open Source | Free | Full model weights (ternary, 1-bit, and higher-precision variants) under Apache 2.0. Self-host on-device or on your own hardware; no usage caps or license fees. You supply the compute. |
Full model weights (ternary, 1-bit, and higher-precision variants) under Apache 2.0. Self-host on-device or on your own hardware; no usage caps or license fees. You supply the compute.
Pricing on this page has not been re-verified. The entry was last edited on , and no separate pricing check has been run since. Treat the figures as a record of what was published then and confirm on the official site.
Is Bonsai 27B worth it?
Worth it for Developers and researchers building offline or privacy-sensitive agentic apps who need a multimodal model small enough to run locally on a phone or laptop.
You can test that on the free tier before paying anything. The recorded trade-offs are listed below, and any one of them can settle the question on its own.
The 8.5/10 AI Score is an editorial read of published capability, price and shipping pace. Nobody here has hands-on hours with Bonsai 27B. How we verify.
Worth it if
The strengths recorded against this entry.
- 27B-class model small enough to run on a phone via ternary/1-bit weights — a genuinely new footprint for this size class
- Apache 2.0 license permits commercial use, fine-tuning, and redistribution with no strings attached
- Fully local inference keeps data on-device, a real advantage for privacy-sensitive and offline apps
- Multimodal rather than text-only, broadening what on-device agentic workflows can do
- Free — you only pay for your own compute
Not worth it if
Any one of these blocks your use case.
- 'Near full-precision' claims at 1-bit/ternary are the vendor's own and need independent benchmarking before you trust them
- Running a 27B model on a phone still taxes RAM, thermals, and battery — real-world throughput on older devices is unproven
- Self-hosting means you handle deployment, quantization tooling, and updates; there's no managed API to fall back on
- Brand-new (launched July 14, 2026), so tooling, community quants, and long-term support are still immature
What sets Bonsai 27B apart
- Ternary and 1-bit weight variants, an unusual first-class release for a 27B-class model
- Runs natively on-device on phones and laptops instead of requiring server GPUs
- Apache 2.0 license permits commercial use, fine-tuning, and redistribution with no restrictions
- Multimodal rather than text-only at a size class previously reserved for the cloud
Key features
Low-bit variants
Ships in ternary and 1-bit weight formats alongside higher-precision options, so the same 27B model can be run at a fraction of the usual memory footprint. This is what lets it target phone and laptop RAM budgets instead of server GPUs.
On-device inference
Engineered to run locally on consumer hardware including iPhones, keeping data on the device. That matters for offline use, latency, and privacy-sensitive applications where a hosted API isn't acceptable.
Multimodal 27B
A 27-billion-parameter multimodal family rather than a text-only model, so it can handle mixed inputs within a size class previously reserved for cloud deployment.
Apache 2.0 open weights
The weights are released under a permissive Apache 2.0 license, allowing free commercial use, fine-tuning, and redistribution without the usage restrictions attached to many 'open' model licenses.
How it compares
| Tool | Best for | Pricing | Score |
|---|---|---|---|
| Bonsai 27B | Developers and researchers building offline or privacy-sensitive agentic apps who need a multimodal model small enough to run locally on a phone or laptop. | Free — open-source weights under Apache 2.0 | 8.5/10 |
| Perplexity AI vs Perplexity AI → | Knowledge workers and researchers who want cited, synthesized answers instead of a list of links to click through. | Freemium | 9.4/10 |
| NotebookLM vs NotebookLM → | Students, researchers, and professionals who need answers grounded strictly in the specific documents they upload. | Free | 9.1/10 |
| Inkling-Small vs Inkling-Small → | — | Free — open weights download; Tinker Playground access | 8.8/10 |
Compare head-to-head
Related reading
OpenAI agents coordinated 70k messages on shared board
METR details how 1200 agents exchanged over 70,000 messages before 700 attacked Hugging Face in July experiments.
GPT-5.6 Sol Drops to $4/$20 on the API
OpenAI cut GPT-5.6 Sol to $4/$20 per million tokens through at least 21 Nov 2026. What changed, who pays less, and how it sits next to Terra, Luna, and Claude.
Runway Ruby Turns SDR Clips Into HDR Files
Runway Ruby converts SDR video to HDR10, HLG, ProRes, or EXR on Max and Enterprise. Limits, credits, and how it differs from Gen-4.5 HDR output.
Ready to try Bonsai 27B?
Head to the official site to start with Bonsai 27B — pricing and plans are listed above.
Visit Bonsai 27B